Written and maintained by CASRAI Editorial Board
Last updated
Neuropixels is a family of silicon probes for recording electrical activity from many neurons at once in the brain of a living animal. Each probe has 384 recording channels, addresses a larger set of recording sites (960 in version 1.0) along its shank, and digitizes the signal on the probe itself. This guide is for neuroscientists and lab managers deciding what the technology is, how the two published generations differ, and what a lab has to plan for. It is research-use background, not a purchasing or experimental protocol.
Short answer
Neuropixels probes were introduced in a 2017 Nature paper as fully integrated silicon probes: a thin shank carrying the recording sites, with amplification, filtering, multiplexing and digitization built into the probe base, so data leave the probe as digital signals. Neuropixels 2.0 (Science, 2021) shrank the probe and headstage to about one third of the earlier size and added multi-shank designs. The key concept to get right is the difference between sites (physical electrodes) and channels (simultaneous recordings).
Definition and scope
Jun et al. describe a probe with a single 10 mm long shank (70 x 20 µm cross-section) and a 6 x 9 mm base. The shank carries 960 recording sites, and 384 recording channels can be programmably connected to them. The signal is filtered, amplified, multiplexed and digitized on the base, which removes the need for an external amplifier stage. Neuropixels belongs to the broader family of multi-channel extracellular recording tools; for the in vitro planar-array relative, see multi-electrode arrays.
How it is used in practice
In the 2017 paper, two probes recorded 741 putative single neurons from five brain structures in an awake, head-fixed mouse. Chronic implants in rats and mice were also reported: 127 neurons including 22 grid cells in rat entorhinal cortex on day 49, stable recordings over more than eight weeks in rat medial prefrontal cortex, and more than 100 neurons recorded over more than 150 days in head-fixed mice.
Each probe records two frequency bands: an action-potential (AP) band sampled at 30 kHz and a local field potential band sampled at 2.5 kHz. The reported gain is selectable between 50 and 2,500, with noise of about 5 µV RMS in the AP band and about 9 µV RMS in the LFP band. The slower band is the one analysed in work on local field potentials; the fast band is what is used to isolate single neurons.
Neuropixels 1.0 versus 2.0
The table compares only values stated in the two original papers.
| Feature | Neuropixels 1.0 (Jun et al., 2017) | Neuropixels 2.0 (Steinmetz et al., 2021) |
|---|---|---|
| Recording channels per probe | 384 | 384 |
| Recording sites | 960 on one shank | 4-shank version: 5,120 sites (1,280 per shank) |
| Site spacing | 20 µm | 15 µm |
| Size and weight | Base 6 x 9 mm; shank 10 mm long | Probe and headstage about one third of the earlier size; two probes plus one headstage about 1.1 g; probe alone 0.19 g |
| Probes per headstage | — | Two (768 channels) |
| Long-term stability reported | Stable over more than 8 weeks (rat); more than 150 days (mouse) | Same neurons tracked over more than 2 months; most implants good for 8 weeks or more; one lab recorded for more than 150 days in all three implants, with a maximum of 309 days from implant to recording |
Where it matters for labs
Because digitization happens on the probe, the lab-side questions shift from analogue amplification to acquisition and data handling. Three points follow directly from the papers.
- Two data streams per probe. The AP band (30 kHz) and LFP band (2.5 kHz) are separate streams, so storage and processing plans should account for both. Include them in the data management plan.
- Drift correction is a post-processing step. The 2.0 paper reports that the probe design enables automatic post-hoc correction for brain movement, and that about 93% (±9%) of well-isolated units were tracked across sessions at most 16 days apart, and about 85% (±19%) across sessions 3 to 9 weeks apart.
- Chronic work depends on weight and size. The reduced headstage mass of 2.0 (about 1.1 g for two probes and one headstage) is what the authors link to stable long-term recordings.
Common misunderstandings
“Neuropixels records from 960 neurons at once.”
No. A 1.0 probe has 960 sites but only 384 channels, so at any moment it records from a programmable subset of 384 sites. The site count describes where recording is possible, not how many neurons are recorded.
“More sites means more channels.”
Not in the published designs. The 4-shank 2.0 probe has 5,120 sites but still 384 channels per probe.
“It is one product.”
The two papers describe different generations with different geometry and size, so results and hardware details from one should not be assumed for the other.
FAQ
What is Neuropixels?
A high-density silicon probe for recording neural activity, with 384 recording channels and on-probe digitization, first described by Jun et al. in 2017.
How many channels does a Neuropixels probe have?
384 per probe in both published generations; one 2.0 headstage serves two probes, giving 768 channels.
What changed in Neuropixels 2.0?
A smaller, lighter probe and headstage, multi-shank designs with 15 µm site spacing, and recordings stable enough to track the same neurons for months.
Sources
- Jun JJ et al. Fully integrated silicon probes for high-density recording of neural activity. Nature 2017;551(7679):232-236. PMC5955206. Read 2026-11-11.
- Steinmetz NA et al. Neuropixels 2.0: A miniaturized high-density probe for stable, long-term brain recordings. Science 2021;372(6539):eabf4588, doi:10.1126/science.abf4588. PMC8244810. Read 2026-11-11.
